Identifying key actors in informal collaborative networks in schools

Many network studies focus on bounded, whole networks that capture the interactions between actors working within and across educational institutions such as schools or universities. In many studies of this type, the identification of central actors within the network is associated with one or more research questions (Daly, 2010). Classic node-level centrality measures such as in-degree or out-degree, closeness, or betweenness have long been the staple metrics used to identify these key actors (Freeman, 1979).We know that network data is distinct from many datasets in that it violates the assumption of independence between cases. Techniques within SNA have been developed to adjust for this interdependent nature of the data (Wasserman and Faust, 1994). Despite this heritage, node level centrality measures are often treated as independent data when determining those who are most central to the network activity. Actors are usually ranked by individual centrality measures and those with the highest scores used to determine who plays the major roles within the network. This independent ranking approach does not take into account the interdependent nature of centrality within the network.A key question remains therefore: Do analytical techniques that recognise the interdependence of centrality across different parts of the network yield a different set of central actors than the typical independent ranking approach?Network data were collected from primary and secondary schools in the south of England. Teachers in each school were asked to identify those colleagues to whom they turned for resources and advice in teaching and learning, and in the use of assessment data, yielding multiple collaborative networks per school. The networks were analysed using UCINET (Borgatti et al., 2002) and actor-level centrality measures calculated to rank each teacher in the networks. These rankings were compared to sets of four central actors extracted using Borgatti’s (2010) Key Player algorithms, using two approaches to identify sets of central actors. One maximising the reach across the whole network, the other maximising network fragmentation as each identified key player was removed from the network.The independent ranking approach focused actors clustered around the core of the network. The Key Player algorithms developed by Borgatti (2006) identify some of these same actors but indicate a level of redundancy once one or two of the key players is drawn out from the core part of the network. Actors with centrality in a part of the network away from the core are also extracted in the Key Player sets. This maximises overall network reach or fragmentation. These actors would not be identified using the simple rank approach. This was especially the case in networks of larger schools and in networks with lower network cohesion